The opportunities and challenges of AI in health care

Critics have long faulted U.S. medical education for being hidebound, imperious and out of touch with modern health-care needs. The core structure of medical school—two years of basic science followed by two years of clinical work—has been in place since 1910.

Now a wave of innovation is sweeping through medical schools, much of it aimed at producing young doctors who are better prepared to meet the demands of the nation’s changing health-care system.

When we asked dozens of venture capitalists where they see the most potential for applied artificial intelligence, they unanimously agreed on health care. Technology has already been used to incrementally improve patient medical records, care delivery, diagnostic accuracy, and drug development, but with AI we could achieve exponential breakthroughs.

Deep learning first caught the media’s attention when a team from the lab of Geoffrey Hinton at the University of Toronto won a Merck drug discovery competition despite having no experience with molecular biology and pharmaceutical development. Recently, a multidisciplinary research team at Stanford’s School of Medicine comprised of pathologists, biomedical engineers, geneticists, and computer scientists developed deep learning algorithms that diagnose lung cancer more accurately than human pathologists.

The ultimate dream in health care is to eradicate disease entirely. This dream might be possible one day with the assistance of AI, but we have a very, very long way to go.

Innovation is challenged by risk aversion and digitization

“Health care as a system advocates ‘do no harm’ first and foremost. Not ‘do good,’ but ‘do no harm.’ Every application of AI in healthcare is regulated by that fundamental philosophy,” cautions Kapila Ratnam, a scientist turned venture partner at NewSpring Capital. Additionally, Lisa Suennen, managing director at GE Ventures, highlights that “the single biggest contribution to excess cost and error in healthcare is inertia.” The attitude of “this is how it’s always been done” is literally killing people.

Other investors agree that the ultraconservatism in the health care system, while intended to protect patients, also harms them by restricting innovation. Gavin Teo, partner at B Capital Group and a specialist in digital health, cites “provider conservatism and unwillingness to risk new technology that does not provide immediate fee-for-service (FFS) revenue” as a major challenge for startups tackling health care. Teo also points out that the industry feels burned from recent experiences, such as “electronic medical records (EMR) digitization regulations, which were overhyped and resisted.”

There are many well-known challenges to implementing machine learning and AI in health care. The first is the lack of “curated data sets,” which are required to train AI via supervised learning. “Curated data sets that are robust and have both the breadth and depth for training in a particular application are essential, but frequently hard to access due to privacy concerns, record identification concerns, and HIPAA,” explains Dr. Robert Mittendorff of Norwest Venture Partners.

Summerpal Kahlon, MD, is director of care innovation at Oracle Health Sciences. He’s seen many of these data challenges firsthand in delivering technological infrastructure to support individualized care. “Adverse drug events cause around 770,000 injuries and deaths annually in the U.S. and cost each hospital up to $5.6 million annually,” Kahlon discloses, “but drug data is messy, coming from multiple sources in multiple formats. Additionally, genetic data in support of pharmacogenomics is not available at scale yet.”

Fixing accidental hospital infections and performing rare disease detection with AI also requires better data than is currently available. According to Kahlon, the genetic and behavioral data required for rare disease studies are not “well-defined nor easily captured,” while “much of the information relating to the risk factors for hospital-acquired infections is kept in unstructured notes in the chart, including in flowsheets and clinical notes.”

While data problems in health care abound, another major challenge is designing technical solutions that can be smoothly implemented and integrated into clinician practices and patient care. “Behavioral change is the blockbuster drug of digital health,” claims Dr. Mittendorff, but changing habits is much easier said than done. The wrong solution or rollout can even harm the health care industry.

Implementing and integrating technology has indeed been a burden for many clinicians and practitioners. Dr. Jose I. Almeida is a pioneer in endovascular venous surgery who has practiced for over 20 years. He adopted electronic health records (EHR) ahead of the curve, yet has not seen many of the promised benefits. “We implemented our first EMR System eight years ago hoping it would improve efficiencies. We are now on our fourth system, and remain disappointed,” complains Dr. Almeida. “Right now, it’s been more of a hassle than a time-saver, and has actually disrupted the doctor/patient relationship by forcing a screen between physicians and their patients.”

Leonard D’Avolio, founder of Cyft, has harsh feedback for fellow entrepreneurs trying to tackle the space: “We’re seeing hospital after hospital take incredible loss and have widespread layoffs simply from the challenge of implementing electronic health records. Imagine what happens if you then show up and say ‘I have artificial intelligence’.”

The health care industry is just getting its arms around capturing data digitally, yet many tech entrepreneurs mistakenly believe that creating a dashboard or dropping in a product will somehow lead to adoption of technology and improve operations. “There’s a huge misconception that AI requires huge amounts of data, but that’s not the real issue in health care. The real issue is understanding the context into which you are trying to introduce these technology,” warns D’Avolio. “You need context and a deep understanding of who will use this. What workflows will be introduced?”

Even if a medical provider does successfully digitize their data, technical carelessness can introduce problems for everyone in the system. According to Ratnam of NewSpring, “A credit card record costs about 10 cents on the black market. A medical record costs about $200. Medical data is so valuable that hackers constantly seek ways to break into provider or payment systems and other repositories of medical data.”

There is often tension between a venture-backed company, which aims for fast growth, and the health care system, which challenges scale because of environmental complexity and unavoidable hand-holding.

“This lesson has not been widely learned,” observes D’Avolio.

…But opportunities abound and solutions exist

Despite challenges, innovation in health care must continue. According to Teo of B Capital, “A study by the Association of American Medical Colleges estimates that by 2025 there will be a shortfall of between 14,900 and 35,600 primary care physicians.” At the same time, the population is aging and in need of more medical attention.

Thus, inaction and failure to innovate may lead to doing harm.

Luckily, many companies strive to address these issues before they come to pass. CB Insights recently profiled 106 different artificial intelligence startups in health care tackling the various challenges in the space, ranging from patient monitoring to hospital operations.

Teo identifies AI powered chatbots and virtual assistants as one way to “alleviate supply constraints by widening the reach of video telehealth options. In this case, diagnosis can be powered by machine learning and then trained by artificial intelligence.” Examples of companies providing clinician assistant and care delivery services include Babylon Health, Evidation Health, Sensely, and Seniorlink.

Artificial intelligence can not only improve care delivery, but also assist in clinician decision-making and operational efficiency, amplifying the impact of each individual practitioner. AnalyticsMD employs AI and machine learning (ML) to streamline hospital operations in emergency rooms, operating rooms, and inpatient wards, while predictive companies like Cyft and HealthReveal analyze disparate data sources to accurately triage and apply interventions to the highest risk patients.

AI helps not only physicians, but also patients. A study by the Mayo Clinic determined that 50 percent of patients have difficulty with medication adherence. Companies like AI Cure employ computer vision techniques to enable smartphones to recognize faces and medications, lowering the cost and improving the effectiveness of tracking and adherence programs. According to Dr. Mittendorff, “AI enabled coaching will allow a provider or coach to manage more than 1,000 patients simultaneously rather than 50-100, a 10x increase in labor leverage.”

Finally, drug discovery companies like NuMedii and Kyan Therapeutics reduce risk in the drug development process, enabling “powerful and proprietary new combination therapies, as well as individualized treatment with unprecedented efficacy and safety,” according to Teo. Otherwise, Suennen points out that the “general spend for each drug brought to market is $2.5 billion.”

Even technology challenges that come with digitizations can be mitigated by AI. Remember how valuable medical records are to hackers? Many of these records are pilfered through social engineering methods, such as phishing or fraudulent phone calls. Protenus is a health care security company that applies AI to analyze enterprise-wide access logs and flag suspicious cases for administrator review.

Aligning with policy and revenue is a key to success

The key to adoption of health care IT is to identify the correct point of entry and fit these systems seamlessly into existing workflows. D’Avolio of Cyft has spent over 12 years fitting ML into the health care system, yet when he speaks at conferences for clinicians, he avoids using the words “artificial intelligence” or “machine learning” and instead focuses on real impact and benefits.

Many patients with chronic diseases like diabetes visit doctors and hospitals numerous times, costing themselves, insurance providers, and the medical system a substantial amount of money. Cyft builds sophisticated models that identify patients with a preventable readmission and matches them to appropriate intervention programs. Traditionally, these decisions are made by looking at 7 to 10 administrative variables, but Cyft’s model looks at over 400 data sources, ranging from free-text input from nurses to call center data. While adoption of such technologies may seem complicated, D’Avolio gets buy-in by strategically aligning with revenue incentives and policy decisions.

“In healthcare, policy eats strategy and culture for breakfast,” explains D’Avolio. “For example, prior to [when] the American Recovery and Reinvestment Act passed in 2009, the rate of adoption of electronic health records was under 9%. Today, thanks to the carrot and stick incentives involved in that act the rate of adoption is > 90%.” Another major policy shift that has dramatically helped investment in health care IT is the value-based care experiments (also called demonstration programs) funded by the Center for Medicare & Medicaid Innovation (CMMI).

Knowing which policy an organization is incentivized or paid by is key to identifying promising customers. According to D’Avolio, “Organizations that get paid mostly from seeing more patients will want AI that helps deliver more complex care faster. Organizations that are paid via value-based programs will seek technology that keep patients healthier at lower cost.”

Suennen of GE Ventures agrees that operational analytics can dramatically improve health systems: “25 percent of the more than $7 billion spent each year on knee and hip surgeries are impacted by bundled payments initiatives. Determining how to manage these bundles is challenging, and advanced technologies can aid in understanding what changes must be made across the board in operations and financial/clinical management to ensure that health systems can respond.”

Teo is also excited by policy changes that should drive forward health care innovation. “New reimbursement driven by the Medicare Access and CHIP Reauthorization Act (MACRA) and the Merit-based Incentive Payment System (MIPS) incentives in 2017 will drive quality outcomes, phasing providers to think more holistically when investing in technology.” Additionally, he believes that a looser FDA in the coming years will help drive investment in personalized medicine.

Successful health care innovation will only happen with strong collaboration between entrepreneurs, investors, health care providers, patients, and policy developers. If the stars align, humanity stands to derive enormous benefit from the application of AI and inch closer to our dream of perfect health and a world without disease.

This article appeared originally at TopBots.